Diagnosing tuberculosis in the absence of bacterial excretion is a complex and ambiguous task even for phthisiatricians with extensive experience. This article examines 139 input parameters taken into account when diagnosing tuberculosis. Managing such a number of characteristics while diagnosing is challenging even for professionals. Therefore, during the human-machine procedure, an assessment of the informativeness of the features is made and their number is reduced to 62. After this, professional phthisiatricians divide them into six groups based on their own experience and regulatory requirements. Classification tasks are solved for each group using seven classical machine learning methods. The results of the division into groups and the quality metrics of various methods are presented in tables. The article is of interest both for professional phthisiatricians and specialists in machine learning.